Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

HL: Fachverband Halbleiterphysik

HL 2: Focus Session: Biocompatible Organic Semiconductors for Artificial Intelligence

HL 2.7: Vortrag

Montag, 9. März 2026, 12:30–12:45, POT/0051

Redundant information in physical reservoir computing — •Andreas Hofacker, Richard Kantelberg, Hans Kleemann, and Karl Leo — Dresden Integrated Center for Applied Physics and Photonic Materials (DC-IAPP), TU Dresden, Dresden, Germany

Maximizing the processing power of a physical reservoir given its physical constraints is crucial for practical applications, but remains an open problem. As a basis for such an optimization, a quantity measuring reservoir capability is needed. To address this need, we propose the use of independent component analysis for assessing information content in reservoir outputs. We present evaluations of organic mixed ionic-electronic conductor based physical and simulated reservoirs and show that task-specific performance is linked to information redundancy in the output channels. By leveraging this insight, sparser reservoir readout can be realised without loss of performance.

Keywords: Neuromorphic computing; Reservoir computing; Organic mixed ionic-electronic conductors; Information theory